Trans fatty acids (tFAs) are formed during the partial hydrogenation of vegetable oils, and there has been speculation about the effect of their consumption on cancer risk. 1 Mechanisms through which tFAs could cause cancer include disruption of the phospholipid cell membrane and associated enzymes and receptors, 2–4 and disruption of eicosanoid synthesis. 3,4 These potential mechanisms provided a rationale for looking at the relation between colorectal adenomas and consumption of foods containing partially hydrogenated vegetable oil (PHVO). We were interested in the effects of actual whole food sources of tFA—not tFA, per se—because it is difficult to estimate a single nutrient effect using observational dietary data. One can never be certain an observed association is attributable to the nutrient of interest or to other simultaneously present constituents of whole foods. Furthermore, actual dietary exposure is to whole foods, not nutrients in isolation. Estimating the effects of consuming whole foods has the advantage of providing direct information on how one could modify diet to reduce risk of disease. PHVO is the principal source of tFA in the American diet. 5 Meat and dairy from ruminant animals are also sources of dietary tFAs. 5 We limited our investigation to foods containing PHVO.
A previous case-control study looked at the relation between colorectal adenomas and consumption of foods containing PHVO in a Los Angeles sigmoidoscopy screening clinic. 6 Those data demonstrated a twofold increased prevalence of adenomas (95% confidence limits (CL) = 1.3, 3.5) among subjects consuming 350 or more kcals of sweetened baked goods daily, relative to subjects consuming less than 50 kcals daily. There was little evidence of increased prevalence associated with other food groups (oils and condiments, french fries and chips, and candy bars). We report here results from a second case-control study of foods containing PHVO and colorectal adenomatous polyps diagnosed in subjects referred for colonoscopy.
Subjects and Methods
We ascertained a total of 236 cases with adenomatous polyps and 407 controls among patients referred for colonoscopy at the University of North Carolina Hospitals in Chapel Hill, between July 1988 and March 1990. Men and women 30–89 years of age were eligible for study. Cases were those diagnosed with at least one adenomatous polyp. Subjects with no polyps, or hyperplastic polyps only (N = 46), were classified as controls. A random sample of polyp specimens (N = 28) resubmitted to a pathologist in a blind fashion indicated 95% reliability in diagnosing adenomatous (case) vs nonadenomatous (control) polyps.
Subjects were ineligible if they had any of the following conditions: (1) history of polyposis (>100 polyps), (2) colitis of any type (eg radiation induced, infectious, idiopathic), (3) previous bowel surgery, (4) previous adenomatous polyp, (5) previous colon cancer, (6) unsatisfactory colon preparation as judged by the colonoscopist, (7) incomplete examination (eg cecum not reached for a reason other than obstruction due to presence of cancer). We instituted the first five of these eligibility requirements to eliminate subjects with conditions that could both influence discovery of adenomas and induce dietary changes such that reported diet was not etiologically relevant exposure. Out of 2,094 colonoscopies performed, 936 subjects (45%) met eligibility criteria. Among eligible subjects, a total of 161 (17%) refused to be interviewed, and interview data were either incomplete or subjectively deemed unreliable by the interviewer for another 130 (14%). We assigned case or control status after interview data had been collected.
Dietary and Covariate Measurement
A graduate nutritionist who was blind to case-control status interviewed eligible subjects over the telephone. Diet during the year before colonoscopy was measured using the 1987 long version 2.1 of the Block quantitative food frequency questionnaire (FFQ). 7,8 In addition to being asked the frequency of consumption for 98 food items, subjects were asked to estimate usual portion sizes. Nutrient data were analyzed using Dietsys V. 3.6 software developed by the National Cancer Institute. 9 Subjects whose daily caloric intake fell below 500 kcals were excluded (2 cases; 2 controls), leaving 234 cases and 407 controls for analysis.
The interviewer also obtained information on potential nondietary confounders: smoking history, physical activity, family history of colorectal cancer, alcohol consumption, body mass index (BMI), and demographics. Subjects who reported having smoked more then 100 cigarettes in their lifetime were classified as “ever smokers”; those who were no longer smoking were classified as “past smokers.” We created a physical activity index using a modified version of a scale developed by Baecke that rates subjects according to frequency and intensity of activity. 10,11
We selected foods containing PHVO from the FFQ based on a supermarket review of ingredients listed on package labels. Because vegetable oils become partially hydrogenated during high temperature frying, we included french fries, chips, and the oil used for frying fish and chicken among the selected foods. We then grouped foods as in a previous study, which addressed an identical research question (Table 1), so we could compare results. The food items in each group contain similar ingredients and modes of preparation. We defined four food groups: (1) Sweetened Baked Goods - principal ingredients are refined flour, sugar and fat; (2) Chocolate Candy - principal ingredients are sugar and fat; (3) Oils and Condiments - principal ingredient is fat; and (4) French-Fries and Chips - principal ingredients are starch and fat. Each item in these food groups contains PHVO as a principal source of its fat content. We calculated daily caloric intake from each food group for each study subject. Because deep-frying chicken or fish produces PHVO, we subtracted the fat content in broiled chicken or fish from the fat content in an equal portion of fried chicken or fish and included the remaining kcals with the oils and condiments grouping. Our nutrient database contained no information on actual tFA content of any of the food items. We measured red meat and vegetable consumption by summing the daily energy intakes of the questionnaire food items in each category.
We used unconditional logistic regression to estimate crude and adjusted odds ratios for adenomas according to 100 kcal increments of daily consumption from each of the main exposures. If the range of consumption was limited, we used 50 kcal increments. We did not expect category increases of less than 50 kcals/day to produce an epidemiologically detectable change in disease risk, so we chose category cutpoints to be at least 50 kcals/day wide. We examined the influence of the category cutpoints by comparing results from models that used various categorizations. If there were a large number of subjects (50+) consuming nothing from a food group, we included a “0 kcal” category. We specified age, sex, smoking, weight, height, alcohol consumption, physical activity, total energy intake, red meat consumption, vegetable consumption, and family history as potential confounders. All exposure food groups were included in one model. We compared the results obtained after assigning missing height, weight, and physical activity values the median for their respective gender and case-control group with results from analyses that excluded subjects with missing values. We adjusted for continuous covariates using quadratic splines with the upper tails restricted. 12 The spline’s ability to conform to the confounder-disease relation in the data provides more complete control of confounding than conventional approaches.
Table 2 presents characteristics of the study groups. Cases were slightly older, more likely to be male, and more likely to be overweight than controls. They were slightly more likely to have ever smoked and more likely to drink an average of more than 100 kcals of ethanol daily. They were less likely to engage in vigorous physical activity. Reasons for colonoscopy were similar in cases and controls.
Table 3 presents crude and adjusted ORs and 95% CLs for colorectal adenomas according to consumption of foods containing PHVO. Both crude and adjusted data showed a positive association between adenoma prevalence and increasing consumption of sweetened baked goods, although the relation was not strictly increasing. Consumption of 400+ kcals/day was associated with a 1.9-fold increase in the odds of having an adenoma, relative to consuming less than 100 kcals/day (95% CL = 0.95, 3.8). When the uppermost category was defined as 200+ kcals/day, the OR still equaled 1.9 (95% CL = 1.1, 3.1). Consumption of 200+ kcals/day of oils and condiments was associated with a 2.4-fold increase in odds of adenoma, relative to consuming less than 100 kcals/day (95% CL = 1.3, 4.2). Results did not differ materially when category cutpoints were changed. The categorical associations between adenomas and chocolate candy were monotonically decreasing (OR = 0.76 and 0.49 for >0–<50, and 50+ kcals/day vs 0 kcals/day, respectively), while the associations with fries and chips went up and then down. We looked at ORs comparing those who ate any amount of fries and chips with those who ate none to eliminate the unlikely nonmonotonic trend observed when the data were more finely categorized. The corresponding OR for fries and chips was 1.4 (95% CL = 0.85, 2.2). None of these results differed materially when subjects with missing covariate data were excluded from the analyses (61 cases and 58 controls, or 19% of all subjects).
The approximate doubling of the prevalence odds observed among subjects who consumed 200 kcal or more per day of sweetened baked goods is comparable to the association observed in a similar study conducted in a Los Angeles sigmoidoscopy screening clinic. 6 Neither the positive association seen for oils and condiments nor the negative association seen for chocolate candy was present in the first study. Results for fries and chips in this study were not clearly negative or positive. Inconsistencies between the two studies might be due to variation in specific foods consumed by each population, to differences in uncontrolled sources of bias, to the use of colonoscopy-defined cases in the North Carolina study, or to random variation.
An advantage in using colonoscopy for classifying subjects on disease status is reduction in the likelihood of disease misclassification, since subjects are examined over the entire length of the colon. A disadvantage in using a colonoscopy-referred population is that controls may not be representative of the disease-free population that gave rise to cases. Cases and controls experienced identical selection factors, however, and had very similar distributions for number of hospitalizations in the past 5 years (results not shown), as well as having similar indications for colonoscopy (Table 2).
A weakness of the FFQ used for this study is its inability to distinguish food brands that do not contain PHVO. It is inevitable that some reported foods were assumed to contain PHVO when in fact they did not. This possibility has become more likely over time owing to the introduction into the marketplace of foods characterized by alternative ingredients and preparation methods. Consequently, it may not be valid to extrapolate results from this study to dietary exposures today.
For the prevalence odds ratios in this study to approximate incidence odds ratios, the exposure food groups must not be associated with adenoma duration. If a food exposure was positively associated with progression of an adenoma to colorectal cancer (CRC), it could appear negatively associated with prevalent adenomas. Similarly, a food exposure may be positively associated with prevalent polyps if it is negatively associated with colorectal cancer (CRC). We have no data to evaluate these possible sources of bias.
Total dietary tFA data were not available for this study, so it was not possible to investigate the tFA-adenoma association. Even if these data had been available, single nutrient associations are inevitably confounded by known and unknown co-constituents in whole food. For example, there are many fatty acid by-products of the hydrogenation process, and a particular tFA association may only reflect an association between one or more other fatty acid by-products and adenoma development. Sweetened baked goods are characterized by high sugar, high fat, low fiber, and low mineral and micronutrient content. Any or all of these characteristics may be responsible for an effect on colorectal adenomas as well. In the Los Angeles sigmoidoscopy study, a positive association between dietary tFA and colorectal adenomas was removed upon adjustment for sweetened baked goods and other confounders (OR = 0.90 for 6+vs <2 g/day; 95% CL = 0.40, 2.0). An advantage of estimating whole-food effects is that they provide direct information for formulating dietary guidelines. Dietary recommendations do not necessarily follow from knowledge of a nutrient’s effect. For example, liver contains folic acid, but its effect on health is not likely to be the same as the effect of a serving of Brussels sprouts that supplied an equivalent amount of folic acid.
This study was conducted to determine whether results from the Los Angeles sigmoidoscopy study could be corroborated. We considered evidence for an association between consumption from a food group and adenomas to be strongest when we found consistency between our two studies. We conclude that it is worth further investigating the relation between sweetened baked goods and colorectal adenomas based on the consistency between results across studies. The lack of consistently positive associations between adenomas and other PHVO-containing food groups across the studies conflicts with the hypothesis that PHVO is responsible for the sweetened baked goods association, although our results may only reflect large error in measuring consumption of PHVO. It is also possible that consumption of sweetened baked goods is acting as a marker for another dietary factor that we measured poorly or not at all. Nonetheless, we know of no unmeasured adenoma risk factor that would be capable of generating a spurious association as large as seen in our two studies, since the effects of uncontrolled confounders must be considerably larger than the spurious associations they produce. 13
We thank Sharon Murray for her help with data management and Corinne Aragaki for her helpful comments on the manuscript.
1. Kummerow FA. Dietary effects of trans fatty acids
. J Environ Pathol Toxicol Oncol 1986; 6:123–149.
2. Beare-Rogers JL. Trans- and positional isomers of common fatty acids. Adv Nutr Res 1983; 5:171–200.
3. Kinsella JE, Bruckner G, Mai J, Shimp J. Metabolism of trans fatty acids
with emphasis on the effects of trans, trans-octadecadienoate on lipid composition, essential fatty acid, and prostaglandins: an overview. Am J Clin Nutr 1981; 34:2307–2318.
4. Dwyer J. Human studies on the effects of fatty acids on cancer: summary, gaps and future research. Am J Clin Nutr 1997; 66(suppl):1581s–1586s.
5. Emken EA. Nutrition
and biochemistry of trans and positional fatty acid isomers in hydrogenated oils. Ann Rev Nutr 1984; 4:339–376.
6. McKelvey W, Greenland S, Chen MJ, Longnecker MP, Frankl HD, Lee ER, Haile RW. A case-control study of colorectal adenomatous polyps
and consumption of foods containing partially hydrogenated oils
. Cancer Epidemiol Biomarkers Prev 1999; 8:519–524.
7. Block G, Hartman AM, Dresser CM, Carroll MD, Gannon J, Gardner L. A data-based approach to diet
questionnaire design and testing. Am J Epidemiol 1986; 124:453–469.
8. Block G. Health habits and history questionnaire: Diet
history and other risk factors. Personal Computer System Packet, Division of Cancer Prevention and Control, National Cancer Institute, Bethesda, MD, 1989.
9. NCI Information Management Services, Inc, and Block Dietary Data Systems. DIETSYS Version 3.0 User’s Guide. January 9, 1994.
10. Baecke JAH, Burema J, Frijters JER. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr 1993; 36:936–942.
11. Sandler RS, Pritchard ML, Bangdiwala SI. Physical activity and the risk of colorectal adenomas. Epidemiology 1995; 6:602–606.
12. Greenland S. Dose-response and trend analysis in epidemiology: alternative to categorical analysis. Epidemiology 1995; 6 (4):356–365.
13. Breslow NE, Day NE. Statistical Methods in Cancer Research. vol. 1. The Analysis of Case-Control Studies. IARC Scientific Pub. No. 32. Lyon: International Agency for Research on Cancer, 1980; 192–246.